search

e0404 / matRad

An open source multi-modality radiation treatment planning sytem developed by e0404 @ DKFZ
http://www.matRad.org
Other
223 stars 176 forks source link

optimization sanity checks #404

Closed regproj closed 4 years ago

regproj commented 4 years ago

I just had a couple quick questions. The first question was I obtained this plot of the objectivve function after running fluence optimization on a prostate case. It seems like the final loss is not necessarily the lowest one. I've attached the figure and am wondering if it looks normal.

The second question was is there some way to check whether the optimization constraints are infeasible or not? loss

markbangert commented 4 years ago

Is this the result of the direct aperture optimization following initial fluence optimization?

regproj notifications@github.com schrieb am Mi., 4. März 2020, 07:12:

I just had a couple quick questions. The first question was I obtained this plot of the objectivve function after running fluence optimization on a prostate case. It seems like the final loss is not necessarily the lowest one. I've attached the figure and am wondering if it looks normal.

The second question was is there some way to check whether the optimization constraints are infeasible or not? [image: loss] https://user-images.githubusercontent.com/53584374/75850353-f631e300-5d9b-11ea-897f-a9d8b1b13458.jpg

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/e0404/matRad/issues/404?email_source=notifications&email_token=ACNPY7NNELFU27NE6RGNRB3RFXWLTA5CNFSM4LA4TY42YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4ISIIZKQ, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACNPY7MYOCSGSMGTZDHINETRFXWLTANCNFSM4LA4TY4Q .

regproj commented 4 years ago

this is following initial fluence optimization

markbangert commented 4 years ago

regarding your first question: things look normal. the bahaviour may be related to the interior point optimizer itself and the characteristics of the DAO, i.e., the non-convexity and this discontinuity in the threestep optimization process (initial fluence optimization, sequencing, DAO).

regarding your second question: no.

wahln commented 4 years ago

Just some addition to the comments: This looks to me like an optimization with a lot of constraints. There such jumps can happen due to adaptation of the barrier I think, but as Mark said, there is no safe way to check feasibility upfront. If you want some more output to see plots of constraint violation etc during optimization, you can also try to use the fmincon interface (if you have the optimization toolbox).